US Accuses China of Industrial-Scale AI Theft

White House Accuses China of 'Industrial-Scale' AI Theft Weeks Before Trump-Xi Summit

The White House Office of Science and Technology Policy (OSTP) has formally accused Chinese entities of conducting deliberate, large-scale campaigns to steal intellectual property from American artificial intelligence companies — an explosive charge that lands just three weeks before a scheduled summit between President Trump and Chinese President Xi Jinping in Beijing on May 14, 2026.

The accusation, delivered via a memo written by OSTP Director Michael Kratsios and distributed to multiple U.S. government agencies, centers on a technique known as AI model distillation — a process by which outputs from large, advanced AI models are used to train smaller, cheaper competing models. According to the memo, Chinese actors have allegedly been running this process at scale, using tens of thousands of proxy accounts and jailbreaking techniques to evade detection and extract proprietary knowledge from frontier American AI systems.

China has rejected the accusations outright, calling them "pure slander" and "baseless narratives." The dispute is now injecting fresh diplomatic tension into an already fraught geopolitical relationship, with AI, semiconductor export controls, and intellectual property set to dominate the upcoming presidential summit.

What the OSTP Memo Says: 'Industrial-Scale Distillation Campaigns'

The Kratsios memo, first reported by the Financial Times on April 23, 2026, and subsequently confirmed by Reuters, CNBC, Fox Business, and others, is direct in its language. "The United States government has information indicating that foreign entities, principally based in China, are engaged in deliberate, industrial-scale campaigns to distill US frontier AI systems," Kratsios wrote in the memo.

Kratsios also posted publicly about the accusations on X, stating: "The U.S. has evidence that foreign entities, primarily in China, are running industrial-scale distillation campaigns to steal American AI. We will be taking action to protect American innovation."

The memo goes beyond the question of competitive advantage to raise national security concerns. "These distillation campaigns also allow those actors to deliberately strip security protocols from the resulting models and undo mechanisms that ensure those AI models are ideologically neutral and truth-seeking," Kratsios wrote — a warning that the theft could yield AI systems stripped of safety guardrails built into the original American models.

In response to the threat, the White House said it will "explore a range of measures to hold foreign actors accountable," including sharing intelligence with U.S. AI companies about the tactics being employed and the actors involved.

Anthropic's February Disclosure: 24,000 Fake Accounts and 16 Million Exchanges

The OSTP memo did not emerge in a vacuum. U.S. AI companies had already begun raising alarms about what appeared to be coordinated distillation campaigns targeting their systems.

In February 2026, Anthropic disclosed that it had found approximately 24,000 fraudulent accounts used by Chinese AI labs — specifically linked to DeepSeek, Moonshot AI, and MiniMax — to generate 16 million exchanges with its Claude model, all in violation of Anthropic's terms of service. OpenAI has also reported large-scale distillation attempts targeting its systems, with similar links to Chinese AI firms.

Anthropic described the activity as escalating. "These campaigns are growing in intensity and sophistication," the company stated.

The White House memo echoed this assessment, alleging that Chinese actors used "tens of thousands of proxy accounts" to carry out their distillation campaigns at scale — a figure consistent with Anthropic's own findings.

The distillation technique is particularly concerning to U.S. officials and industry because it allows a foreign competitor to capture much of the value embedded in a frontier AI model — value that may represent years of research, billions of dollars in compute, and proprietary training data — at a fraction of the cost of building such a model from scratch. "Industrial distillation activities that aim to systematically undermine American research and development and access proprietary information, however, are unacceptable," Kratsios said in a statement to CNBC.

Congress Moves: The Deterring American AI Model Theft Act of 2026

The OSTP memo is not the only escalation on the U.S. side. On April 15, 2026 — just over a week before the memo became public — Rep. Bill Huizenga (R-MI) introduced H.R. 8283, the Deterring American AI Model Theft Act of 2026, into Congress. The bill was referred to the House Committee on Foreign Affairs and was co-sponsored by Rep. John Moolenaar, Chairman of the House Select Committee on the Chinese Communist Party.

The legislation would authorize sanctions and Entity List designations against Chinese and Russian entities found to be using unauthorized "query-and-copy" techniques — another term for the distillation-based model extraction described in the OSTP memo — on American AI models.

"Model extraction attacks are the latest frontier of Chinese economic coercion and theft of US intellectual property," Rep. Huizenga said in a statement cited by PYMNTS and Bloomberg.

The bill's introduction signals that U.S. lawmakers are treating AI model theft not merely as a commercial dispute, but as a matter of national economic security — one that may soon carry concrete penalties for the companies and governments involved.

China Pushes Back: 'Baseless Narratives' and 'Pure Slander'

China's government has responded with categorical denials at both the diplomatic and official levels.

Chinese embassy spokesman Liu Pengyu called the U.S. accusations "pure slander," according to The American Conservative. The following day, at the Chinese Foreign Ministry's regular press conference on April 22, 2026, spokesperson Guo Jiakun was similarly dismissive. "The allegations of intellectual property theft by China are nothing but baseless narratives," Guo said, adding that China is one of the world's leading countries in intellectual property protection and the top filer of international patent applications.

In support of that position, Guo cited figures from the Foreign Ministry's official record: China claims to have established 81 national-level IPR protection centers, 50 fast IPR protection service centers, and 99 overseas IPR dispute response guidance platforms.

The sharp divergence between Washington's specific technical allegations — citing fraudulent accounts, jailbreaking techniques, and named Chinese firms — and Beijing's categorical denial of any wrongdoing reflects the depth of the dispute. Neither side has indicated any willingness to move toward the other's position ahead of the summit.

Why This Matters: AI, Trade, and the Summit Stakes

The timing of the OSTP memo is difficult to separate from the diplomatic calendar. The Trump-Xi summit was originally scheduled for the end of March before being postponed to May 14. The Kratsios memo dropped three weeks before that rescheduled meeting — a meeting where AI, semiconductor export controls, and intellectual property were already on the agenda.

The allegations add substantial new weight to those discussions. Earlier in 2026, the Trump administration approved conditional Nvidia chip sales to China, though officials indicated that shipments had not yet begun. How the U.S. handles those sales — and whether the new AI theft accusations accelerate or complicate any sanctions or export restrictions — will likely be a central question when Trump and Xi meet in Beijing next month.

The broader context also includes a criminal dimension. In 2024, prosecutors charged a former Google engineer with stealing AI trade secrets and sharing them with Chinese companies — a case that illustrated, at the individual level, the same pattern of alleged IP extraction that the White House is now describing at a state-sponsored, industrial scale.

For the AI industry, the implications are significant. If the U.S. government begins actively sharing threat intelligence with American AI companies — as the OSTP memo suggests it will — that could represent a meaningful shift in how companies like OpenAI, Anthropic, and Google defend their systems. It may also accelerate the adoption of technical countermeasures against distillation attacks, and could affect how AI companies design their APIs and rate-limit access to model outputs.

Expert Reactions

Michael Kratsios, Director of the White House Office of Science and Technology Policy, was unambiguous in his public statements. "The United States government has information indicating that foreign entities, principally based in China, are engaged in deliberate, industrial-scale campaigns to distill US frontier AI systems," he wrote in the OSTP memo. His comments to CNBC added that activities "aimed to systematically undermine American research and development" were "unacceptable."

On the legislative side, Rep. Bill Huizenga framed the threat in stark terms: "Model extraction attacks are the latest frontier of Chinese economic coercion and theft of US intellectual property."

China's Foreign Ministry spokesperson Guo Jiakun countered that "the allegations of intellectual property theft by China are nothing but baseless narratives," while Chinese embassy spokesman Liu Pengyu called the accusations "pure slander."

Anthropic, for its part, stated that distillation campaigns against its systems were "growing in intensity and sophistication" — a characterization consistent with the OSTP memo's framing of an escalating and organized threat.

What's Next

The immediate focus shifts to the Trump-Xi summit on May 14, 2026, in Beijing. AI theft, semiconductor export controls, and intellectual property are all expected to be on the agenda. Whether the OSTP memo's accusations translate into concrete diplomatic demands — or whether the summit seeks to de-escalate tensions — remains to be seen.

On the legislative side, H.R. 8283, the Deterring American AI Model Theft Act of 2026, has been referred to the House Committee on Foreign Affairs. Its passage would authorize sanctions against Chinese and Russian entities involved in model extraction attacks, though the bill's progress through Congress is not yet determined.

The White House has said it will "explore a range of measures" to hold foreign actors accountable and share intelligence with U.S. AI companies. What those measures look like in practice — and how quickly they are implemented — will depend in part on how the summit unfolds and whether any diplomatic framework emerges for addressing the underlying dispute.

For now, both governments are holding firm to their respective positions: the U.S. characterizing the activity as a state-linked campaign of industrial-scale theft, and China characterizing the allegations as baseless political slander. The gap between those positions, and the stakes of the AI race underlying them, means this dispute is unlikely to resolve quickly.

For more tech news, visit our news section.

What This Means for You

The race to secure advanced AI is not just a geopolitical story — it shapes the tools, platforms, and productivity systems that workers and organizations rely on every day. As AI becomes more deeply embedded in health monitoring, workflow automation, and personal optimization, the security and integrity of those underlying models matters directly to the people using them. Understanding the forces shaping AI development helps you make smarter decisions about the technology you trust. Join the Moccet waitlist to stay ahead of the curve.